Pith. sign in

REVIEW 3 major objections 5 minor 47 references

A Budgeted Multi-Level Monte Carlo Method for Full Field Estimates of Multi-PDE Problems

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A budgeted multi-level Monte Carlo method computes mean and variance fields over the whole spatial domain at essentially the same CPU time as a single scalar quantity of interest, with peak memory close to that of one deterministic solve.

desk verdict Solid memory-constrained BMLMC extension with a real memory bound, but the abstract's CPU-time parity with a single QoI is unproven and the numerics don't support it. read the letter →

arxiv 2506.01644 v2 pith:HAD2TS5G submitted 2025-06-02 math.NA cs.NA

Niklas Baumgarten , Robert Kutri , Robert Scheichl This is my paper
classification math.NAcs.NA MSC 65C0565N3065Y0560H3535R60
keywords multilevelMonteCarlobudgetedMLMCfullfieldestimatesuncertaintyquantificationstochasticPDEsmulti-PDEproblemsmemoryconstraintrandomfields
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper argues that estimating the statistical mean and variance of a PDE solution over the entire spatial domain, rather than just a scalar quantity of interest, does not have to cost extra. It extends budgeted multi-level Monte Carlo (MLMC) with a memory constraint and a sparse multi-index update algorithm that accumulates full-field statistics on the fly, without postprocessing or storing every sample. The authors prove that peak memory stays comparable to that of a single deterministic solve on the finest level, and they demonstrate numerically that the CPU time matches what a single scalar quantity of interest would require. The method is exercised on a chain of coupled PDEs: SPDE-based Gaussian random field sampling, elliptic subsurface flow, and hyperbolic transport.

What carries the argument

The load-bearing object is the sparse multi-index finite element update algorithm built on the multi-sample FE space. For each level $\ell$ and estimation round, a communication split $s$ decides how many processing units are assigned to sample parallelism versus spatial-domain parallelism, with total cell count $|K_{s,\ell}| = |K_{0,0}| 2^{\ell d + s}$. The memory constraint $bc_{\mathrm{Mem}} |K_{0,L}| < \mathrm{Mem}_B$ turns the budget into a bound on the finest level, formula (4.5) sets the maximum number of samples storable per level, and the accumulative second-order-sum identities of Theorem 3.4 carry the on-the-fly statistics. A curried multi-PDE formulation couples SPDE sampling, mixed-form Darcy flow, and discontinuous Galerkin transport on the same distributed mesh.

What would settle it

Track the maximum of $M_{i,\ell}^{\mathrm{Mem}}2^{\ell d}$ across levels from (4.5) for a fixed memory budget on a low-regularity problem: if the maximum occurs below level $L$, or if the sequence is not monotone, the 'without loss of generality' reduction in Theorem 4.1 fails. A direct test is to run the algorithm with a small memory budget forcing $M_{i,L}^{\mathrm{Mem}}=0$ while coarse-level samples are still in memory, and measure actual peak memory: a peak above a small multiple of $|K_{0,L}|$ would falsify the claimed memory bound.

Watch

Extended reading notes

Core claim

The central claim is that the full spatial solution field can be made the estimator's object of interest at no additional resource cost within the budgeted MLMC framework. The authors introduce a communication split index $s$ alongside the discretization level $\ell$; the pair $(s,\ell)$ defines a multi-sample finite element space whose memory footprint scales like $|K_{0,0}|2^{\ell d+s}$. Choosing $s$ through (4.1) and enforcing the memory budget through (4.5) produces a sparse multi-index set, and Theorem 4.1 proves the total peak memory is bounded by $bc_{\mathrm{Mem}} |K_{0,L}|$, the same order as a deterministic solve on the finest mesh. Accumulative online update formulas maintain the mean and variance fields in a single pass, and the numerical study in Section 5.2 indicates that the resulting CPU-time cost is essentially that of a single quantity of interest.

Load-bearing premise

The memory bound in Theorem 4.1 depends on the unproven assumption that the largest dynamic memory allocation over the whole run occurs on the finest level at the stopping moment, and that $M_{i,\ell}^{\mathrm{Mem}}2^{\ell d}$ is non-decreasing in $\ell$; if either fails, the constant in the bound could be larger than stated.

Editorial extensions

If this is right

  • Full-field mean and variance estimates can be produced in a single pass, so no sample solutions need to be stored or postprocessed.
  • The memory footprint of the stochastic, parallel computation is asymptotically the same as the deterministic solve at the finest level, justifying the memory constraint in the knapsack problem.
  • Users do not need to know the solution regularity or the algorithm's memory demand in advance; both are estimated on the fly and fed into the budget allocation.
  • The same distributed data structure can be shared across different finite element discretizations in a multi-PDE chain, so the method is not tied to one PDE type.
  • The lower bound in Theorem 4.3 implies that, in the memory-limited regime, the smallest achievable error is governed by the solution regularity through $(bc_{\mathrm{Mem}}^{-1}\mathrm{Mem}_B)^{-\alpha}$.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same full-field estimator could make stochastic gradient methods for PDE-constrained optimal control practical, since those gradients are random fields over the spatial domain; this is the application the paper names as its original motivation.
  • Because the variance reduction for the full field is weaker than for a scalar quantity of interest ($\beta_v \le \beta_Y$), CPU-time parity may shrink for problems with strong non-linearities or non-smooth solutions; a testable extension would compare QoI-optimized and full-field-optimized sample allocations on such problems.
  • The SPDE sampling step uses $2^d$ solves per realization through Dirichlet-Neumann averaging, so in three spatial dimensions that constant could dominate the resource budget and should be checked before extrapolating the parity claims to $d=3$.
  • The memory bound's proof assumes monotonicity of $M_{i,\ell}^{\mathrm{Mem}}2^{\ell d}$ in $\ell$; if this fails, the constant $bc_{\mathrm{Mem}}$ could be larger than reported even if the asymptotic order survives.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes a budgeted multi-level Monte Carlo (BMLMC) method for computing full-field statistical estimates (mean and variance) of solutions to coupled PDEs with random data, under both CPU-time and memory budgets. It extends the BMLMC framework of Baumgarten et al. [11] by adding a memory constraint, a sparse multi-index FE update algorithm, and a parallel communication protocol to accumulate spatial statistics on the fly. The main theoretical results are a memory bound (Lemma 4.1), a lower bound on the achievable RMSE under the memory constraint (Corollary 4.3), and an upper bound on the variance of QoI increments relative to the full-field sampling error (Lemma 3.1). Numerical experiments on a three-stage multi-PDE problem (SPDE-based Gaussian field sampling, Darcy flow, and linear transport) on 256 cores are presented to support the claims.

Significance. If the central claims hold, the method would be a significant practical advance: it would allow full-field UQ for complex multi-PDE problems with memory footprint comparable to a deterministic solve and CPU time essentially equal to that of a scalar QoI, without a priori knowledge of convergence rates. The paper includes a working implementation in the M++ framework and experiments on a realistic problem chain. The memory analysis and the sparse multi-index update protocol are original and of independent interest. However, the CPU-time parity with a single QoI is not proven and the numerical evidence does not directly support it at equal accuracy; this weakens the headline claim. The self-referential validation of the rates also reduces the strength of the empirical evidence. The algorithmic machinery, including the Welford/Peebay updates and the communication split, is clearly described and is a useful contribution.

major comments (3)
  1. [Abstract and Section 3.2] The abstract claims mathematical proof that full-field estimates come at the same CPU-time cost as a single QoI, but no theorem in the paper establishes this. Lemma 3.1 and Corollary 3.2 only give beta_Y <= beta_v, which implies the full-field sampling error decays no faster than the QoI variance; if the inequality is strict, the classical MLMC cost analysis in Section 3.1 implies that the cost ratio C_eps(full)/C_eps(QoI) diverges as eps->0. The numerical ON/OFF comparison in Section 5.2 optimizes the sample hierarchy for the QoI and measures only the incremental overhead of the full-field accumulation, not the cost of reaching a prescribed RMSE for the field. The claim needs to be either proven under explicit assumptions on the decay rates and constants, or substantially qualified, and the numerical experiment needs to compare full-field and QoI estimates at equal accuracy.
  2. [Section 4.2, Lemma 4.1] The proof of the memory bound contains an unjustified 'without loss of generality' step: it assumes the maximum dynamic memory allocation can be analyzed at level L and at the stopping time, and that the product M_{i,ell}^{Mem} 2^{ell d} is maximized at ell=L. The monotonicity is not proven and is not implied by (4.5); in fact (4.5) together with bC_0,ell ~ c_Mem |K0,ell| suggests the product is roughly level-independent. A rigorous proof would need to bound the max over all ell and all intermediate rounds, e.g., by showing M_{i,ell}^{Mem} 2^{ell d} <= (MemB - bC_max)/c_Mem, which would still yield the conclusion with a slightly larger constant. As written, the proof is incomplete.
  3. [Section 5.2, Figure 9] The validation of Assumption 2.1 is self-referential: the exponents alpha, beta, gamma are estimated online via (3.4) from the same run data that is then presented as confirmation of the assumption. Moreover, the fitted values of beta_v and beta_Y are not reported, although the CPU-time parity claim depends on their equality. The authors should either provide independent verification of the rates (e.g., from separate longer runs or a priori theory) or clearly label the plots as internal estimates, and they should report the actual rate estimates.
minor comments (5)
  1. [Section 3.1] The phrase 'Under Theorem 2.1' should read 'Under Assumption 2.1'; the same mislabeling occurs elsewhere (e.g., 'Theorem 3.3' for Problem 3.3, 'Theorem 2.2 c)' for Remark 2.2(c)).
  2. [Section 3.1, proof of Lemma 3.1] The second inequality writes E[(||v_ell||_V - ||E[v_ell]||_V)^2], but to equal V[||v_ell||] the inner term should be ||v_ell||_V - E[||v_ell||_V]; as written the equality is not justified.
  3. [Algorithm 1 and Section 3.2] The memory constraint check in Algorithm 1 uses |K_{0,L_i}| while (3.6c) uses |K_L|; please unify the notation to avoid ambiguity.
  4. [Section 5.2] The text refers to 'Theorem 3.1' and 'Theorem 3.2' for Lemma 3.1 and Corollary 3.2; update the cross-references to match the numbered statements.
  5. [Figures 9 and 10] The plots would benefit from a description of the boxes, error bars, and horizontal lines; as printed, the legends are insufficient for a reader to verify the claims about memory and time budgets.

Circularity Check

2 steps flagged · score 6.0 of 10

Partial circularity: the CPU-time parity claim is supported by an ON/OFF construction (QoI-optimal hierarchy, piggybacked field update) that enforces cost equality by design, and Assumption 2.1 is 'verified' with rates fitted from the same run data; the memory theorem is independent.

  1. fitted input called prediction [Abstract; Section 5.2, 'Computational Results' (Figure 9).]
    "One central claim in this paper is that the full field estimates, even though they provide less variance reduction than the estimates for QoIs (cf. Theorem 3.2), do not require additional computational resources in terms of memory or computing time. ... To keep both experiments comparable, we optimize the multi-level hierarchy for the estimation of the QoI and compute the full field estimates simultaneously in the ON case."

    The abstract's CPU-time parity claim is numerically supported only by the ON/OFF experiment, which is constructed with the hierarchy (levels L and sample counts) optimized for the scalar QoI while the full-field estimate is accumulated simultaneously with the same samples. Under that construction the ON-case time equals the QoI-optimal time plus a measured update overhead, so the observed 'no significant additional cost' is forced by the experimental design and tests only the marginal update cost, not equal cost at matched RMSE.

  2. other [Section 3.1, Eq. (3.4); Section 5.2, 'Computational Results' (Figure 9).]
    "To increase robustness, the exponent bαu, which is an estimate for αu, is determined by solving a regression problem, using data collected on the fly. ... The top row verifies that Theorem 2.1 is justified by presenting the estimates for bαu, bαQ, bβv, bβY, bγCT, bγMem computed with (3.4)."

    Assumption 2.1 postulates power-law rates α, β, γ for the field and QoI errors. These exponents are least-squares fits (3.4) to the very on-the-fly estimators (∥E_MC[v_ℓ]∥_V, z²_ℓ[v_ℓ], Ĉ^CT_ℓ) that Section 5.2 then presents as 'verification' of the assumption. The same fitted exponents are the inputs to the sample-allocation rule (3.9), the error estimators (3.3)–(3.5), and the complexity bound (4.6), so the empirical support for the assumed rates is generated from the same data under the same assumed functional form; no independent rate (e.g., from known SPDE regularity or a priori FE analysis) is compared.

full rationale

The memory bound (Theorem 4.1/Lemma 4.1) is independent, in-the-paper content: it rests on the cell count identity |K_{s,ℓ}| = |K_{0,0}|2^{ℓd+s} (4.3) and the algorithm's own stopping rule; the proof's 'without loss of generality' substitution of M_{i,L}^{Mem}=0 into the dynamic max term is not justified (the products M_{i,ℓ}^{Mem}2^{ℓd} are roughly level-independent by (4.5)), so the constant may be larger than 2, but the asymptotic O(|K_{0,L}|) order survives; this is a proof gap, not a circular step. The CPU-time parity claim has no theorem: the abstract's 'provide mathematical proof' overstates what Section 4 proves. Its numerical support is the ON/OFF experiment, which is constructed with a QoI-optimal hierarchy and a piggybacked full-field accumulation, so the demonstrated equality of cost is enforced by the experiment design and measures only the update overhead (step 1). The paper's own Figure 10 experiment, by contrast, optimizes with the Bochner-norm rates (3.9) and reports 'convergence rates are weaker for the full field estimates' and (Mem_B)^{-bαu} ≈ 0.0077 ≫ (Mem_B)^{-bαQ} ≈ 0.000007, which shows equal-accuracy full-field estimation demands a deeper, more expensive hierarchy. The verification of Assumption 2.1 is self-referential because its exponents are least-squares fits to the same run data that are then used as both inputs (3.9), (3.3)-(3.5), (4.6) and as evidence (step 2). Self-citations to [11] (Baumgarten et al.) and [35] (Baumgarten's thesis) are load-bearing for the BMLMC framework and the upper bound in (4.6), but they are peer-reviewed, externally validated prior work with stated assumptions that do not include the present results, so they carry independent evidential weight. The Welford accumulation identity (Lemma 3.4) and the variance comparison (Lemma 3.1) are derived in-paper. Net: one headline claim (CPU-time parity) is partially demonstrated by construction and one verification is self-referential, while the memory analysis and update algorithm are independent; overall partial circularity.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

No new physical or mathematical entities are introduced. The communication split index s, the multi-sample FE space, and the MX-FEM label are algorithmic constructs assembled from existing concepts (Welford updates, MLMC, multi-index methods, FEM).

free parameters (4)
  • theta (variance-bias tradeoff)
    User-chosen hyperparameter in Algorithm 1 splitting the error target between sampling error and bias; no sensitivity analysis is provided.
  • eta (error reduction factor)
    User-chosen factor reducing the MSE target after each estimation round; inherited from continuation MLMC [12].
  • initial sample sequence {M_init}_{0,ell} and L0
    Initial guesses required to start BMLMC; the optimal stopping and final estimates can depend on them.
  • online rate estimates (alpha_u, alpha_Q, beta_v, beta_Y, gamma_CT, gamma_Mem) = e.g., alpha_u~0.78, alpha_Q~1.90 in Figure 10
    Estimated by regression (3.4) from the same run; used in sample allocation (3.9) and in interpreting the Theorem 4.3 lower bound.
assumptions (5)
  • domain assumption Assumption 2.1: bias, variance, CPU cost and memory scale as powers of h_ell (rates alpha, beta, gamma) with positive exponents
    Standard MLMC rate assumptions. The algorithm estimates the rates online, but the theory and the claimed optimality of the sample allocation rest on their existence.
  • domain assumption Memory proportional to FE cells: C_{s,ell}^{Mem} <= c_Mem |K_{s,ell}| (4.4)
    Underlies Theorem 4.1. Assumes the parallel FE implementation stores a multi-sample solution with memory linear in the cell count and no hidden dense copies.
  • domain assumption The transfer operator P_{ell-1}^{ell} is an isometry (used in Lemma 3.1 and the telescoping sum)
    Needed for the variance relation between QoI and Bochner norm; the paper defines it as linear isometric but does not prove it for the chosen FE spaces.
  • ad hoc to paper In Theorem 4.1, the maximum dynamic memory allocation can be analyzed at level L at the stopping time (the 'WLOG' step)
    This is an unproved dominance and monotonicity premise; the proof does not show that earlier rounds or coarser levels do not hold larger temporary meshes.
  • domain assumption Standard finite element error estimates for SPDE sampling, mixed Darcy and dG transport from [23,26,45,46,47]
    The rates in Assumption 2.1 rest on these cited results, which the paper does not re-derive.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Budgeted Multi-Level Monte Carlo Method for Full Field Estimates of Multi-PDE Problems." pith.science (2026). https://pith.science/paper/HAD2TS5G

@misc{pith2026250601644,
  author       = {Pith},
  title        = {Pith review of: A Budgeted Multi-Level Monte Carlo Method for Full Field Estimates of Multi-PDE Problems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HAD2TS5G}},
  note         = {Machine review of arXiv:2506.01644}
}
read the original abstract

We present a high-performance budgeted multi-level Monte Carlo method for estimates on the entire spatial domain of multi-PDE problems with random input data. The method is designed to operate optimally within memory and CPU-time constraints and eliminates the need for a priori knowledge of the problem's regularity and the algorithm's potential memory demand. To achieve this, we build on the budgeted multi-level Monte Carlo framework and enhance it with a sparse multi-index update algorithm operating on a dynamically assembled parallel data structure to enable estimates of the full field solution. We demonstrate numerically and provide mathematical proof that this update algorithm allows computing the full spatial domain estimates at the same CPU-time cost as a single quantity of interest, and that the maximum memory usage is similar to the memory demands of the deterministic formulation of the problem despite solving the stochastic formulation in parallel. We apply the method to a sequence of interlinked PDE problems, ranging from a stochastic partial differential equation for sampling random fields that serve as the diffusion coefficient in an elliptic subsurface flow problem, to a hyperbolic PDE describing mass transport in the resulting flux field.

Figures

Figures reproduced from arXiv: 2506.01644 by the authors.

Figure 1
Figure 1. Exemplary estimation round of the full domain update algorithm. Meshes are either held permanently in memory (green), temporarily created (orange) or not initilized due to a high memory demand (red). Each color represents one of four processing units. The black arrows represent the flow of data within the algorithm. solution is represented by (us,ℓ) Ms,ℓ m=1 = Nh Xℓ n=1 µ (m) n ψ (m) n !Ms,ℓ m=1 ∈ Vs,ℓ := Vs,ℓ(Ks,ℓ)… view at source ↗
Figure 2
Figure 2. First sample solution m = 1 on level ℓ = 8 [PITH_FULL_IMAGE:figures/full_fig_p021_2.png] view at source ↗
Figure 3
Figure 3. Second sample solution m = 2 on level ℓ = 8 [PITH_FULL_IMAGE:figures/full_fig_p021_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Estimated mean field on level ℓ = 8 of four parallel samples [PITH_FULL_IMAGE:figures/full_fig_p021_4.png]
Figure 5
Figure 5. Figure 5: Multi-level mean field approximation after the initial estimation round [PITH_FULL_IMAGE:figures/full_fig_p021_5.png]
Figure 6
Figure 6. Figure 6: Multi-level mean field approximation at an intermediate estimation round [PITH_FULL_IMAGE:figures/full_fig_p021_6.png]
Figure 7
Figure 7. Figure 7: Multi-level marginal sample variance after the initial estimation round [PITH_FULL_IMAGE:figures/full_fig_p021_7.png]
Figure 8
Figure 8. Figure 8: Multi-level marginal sample variance at an intermediate estimation round [PITH_FULL_IMAGE:figures/full_fig_p021_8.png]
Figure 9
Figure 9. Figure 9: Implementations with (ON) and without (OFF) full solution up￾date funcitonality [PITH_FULL_IMAGE:figures/full_fig_p023_9.png]
Figure 10
Figure 10. Figure 10: Extension of reserverd computing times, i.e., TB ∈ {1h, 2h, 3h}. in line with Theorem 4.3 where the lower bound of the smallest possible error depends on α. Here, (MemB) −αbu ∼ 512−0.78 ≈ 0, 0077 is bigger than (MemB) −αbQ ∼ 512−1.90 ≈ 0, 000007 and thus, in the memor…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

47 extracted references · 34 canonical work pages

  1. [11]

    A fully parallelized and budgeted multilevel monte carlo method and the application to acoustic waves,

    N. Baumgarten, S. Krumscheid, and C. Wieners, “A fully parallelized and budgeted multilevel monte carlo method and the application to acoustic waves,” SIAM/ASA Journal on Uncertainty Quantification, vol. 12, no. 3, pp. 901–931, 2024

    2024

  2. [1]

    Multilevel Monte Carlo path simulation,

    M. B. Giles, “Multilevel Monte Carlo path simulation,” Oper. Res., vol. 56, no. 3, pp. 607–617, 2008

    2008

  3. [2]

    Multilevel Monte Carlo methods,

    S. Heinrich, “Multilevel Monte Carlo methods,” LSSC, vol. 1, pp. 58–67, 2001

    work page 2001

  4. [3]

    Multilevel Monte Carlo Finite Element Method for elliptic PDEs with stochastic coefficients,

    A. Barth, C. Schwab, and N. Zollinger, “Multilevel Monte Carlo Finite Element Method for elliptic PDEs with stochastic coefficients,” Numerische Mathematik, vol. 119, no. 1, pp. 123–161, 2011

    2011

  5. [4]

    Finite element error analysis of elliptic PDEs with random coefficients and its application to multilevel Monte Carlo methods,

    J. Charrier, R. Scheichl, and A. L. Teckentrup, “Finite element error analysis of elliptic PDEs with random coefficients and its application to multilevel Monte Carlo methods,” SIAM J. Numer. Anal., vol. 51, no. 1, pp. 322–352, 2013

    2013

  6. [5]

    Multilevel Monte Carlo methods and applications to elliptic PDEs with random coefficients,

    K. A. Cliffe, M. B. Giles, R. Scheichl, and A. L. Teckentrup, “Multilevel Monte Carlo methods and applications to elliptic PDEs with random coefficients,” Comput. Vis. Sci., vol. 14, no. 1, pp. 3–15, 2011

    2011

  7. [6]

    Further analysis of multilevel Monte Carlo methods for elliptic PDEs with random coefficients,

    A. L. Teckentrup, R. Scheichl, M. B. Giles, and E. Ullmann, “Further analysis of multilevel Monte Carlo methods for elliptic PDEs with random coefficients,” Numer. Math., vol. 125, no. 3, pp. 569–600, 2013

    2013

  8. [7]

    Multi-level Monte Carlo finite volume methods for nonlinear systems of conservation laws in multi-dimensions,

    S. Mishra, C. Schwab, and J. ˇSukys, “Multi-level Monte Carlo finite volume methods for nonlinear systems of conservation laws in multi-dimensions,” J. Comput. Phys., vol. 231, no. 8, pp. 3365–3388, 2012

    work page 2012

Show all 47 references
  1. [8]

    Multi-level Monte Carlo finite volume methods for uncertainty quantification of acoustic wave propagation in random heterogeneous layered medium,

    S. Mishra, C. Schwab, and J. ˇSukys, “Multi-level Monte Carlo finite volume methods for uncertainty quantification of acoustic wave propagation in random heterogeneous layered medium,” J. Comput. Phys., vol. 312, pp. 192–217, 2016

    2016

  2. [9]

    Convergence analysis of multilevel monte carlo variance estimators and appli- cation for random obstacle problems,

    C. Bierig and A. Chernov, “Convergence analysis of multilevel monte carlo variance estimators and appli- cation for random obstacle problems,” Numerische Mathematik, vol. 130, pp. 579–613, 2015

    2015

  3. [10]

    Estimation of arbitrary order central statistical moments by the multilevel monte carlo method,

    C. Bierig and A. Chernov, “Estimation of arbitrary order central statistical moments by the multilevel monte carlo method,” Stochastics and Partial Differential Equations Analysis and Computations, vol. 4, pp. 3–40, 2016. BUDGETED MULTI-LEVEL MONTE CARLO METHOD 25

    2016

  4. [12]

    A continuation multilevel Monte Carlo algorithm,

    N. Collier, A.-L. Haji-Ali, F. Nobile, E. von Schwerin, and R. Tempone, “A continuation multilevel Monte Carlo algorithm,” BIT, vol. 55, no. 2, pp. 399–432, 2015

    2015

  5. [13]

    Multilevel Monte Carlo methods,

    M. B. Giles, “Multilevel Monte Carlo methods,” Acta Numer., vol. 24, pp. 259–328, 2015

    2015

  6. [14]

    On multilevel best linear unbiased estimators,

    D. Schaden and E. Ullmann, “On multilevel best linear unbiased estimators,” SIAM/ASA Journal on Uncertainty Quantification, vol. 8, no. 2, pp. 601–635, 2020

    2020

  7. [15]

    Survey of multifidelity methods in uncertainty propa- gation, inference, and optimization,

    B. Peherstorfer, K. Willcox, and M. Gunzburger, “Survey of multifidelity methods in uncertainty propa- gation, inference, and optimization,” SIAM Rev., vol. 60, no. 3, pp. 550–591, 2018

    2018

  8. [16]

    Algorithms for computing the sample variance: analysis and recommendations,

    T. F. Chan, G. H. Golub, and R. J. LeVeque, “Algorithms for computing the sample variance: analysis and recommendations,” Amer. Statist., vol. 37, no. 3, pp. 242–247, 1983

    1983

  9. [17]

    Note on a method for calculating corrected sums of squares and products,

    B. P. Welford, “Note on a method for calculating corrected sums of squares and products,” Technometrics, vol. 4, pp. 419–420, 1962

    1962

  10. [18]

    Numerically stable, scalable formulas for parallel and online computation of higher-order multivariate central moments with arbitrary weights,

    P. P´ ebay, T. B. Terriberry, H. Kolla, and J. Bennett, “Numerically stable, scalable formulas for parallel and online computation of higher-order multivariate central moments with arbitrary weights,” Comput. Statist., vol. 31, no. 4, pp. 1305–1325, 2016

    2016

  11. [19]

    Improved efficiency of a multi-index FEM for computational uncer- tainty quantification,

    J. Dick, M. Feischl, and C. Schwab, “Improved efficiency of a multi-index FEM for computational uncer- tainty quantification,” SIAM J. Numer. Anal., vol. 57, no. 4, pp. 1744–1769, 2019

    2019

  12. [20]

    Multi-index Monte Carlo: when sparsity meets sampling,

    A.-L. Haji-Ali, F. Nobile, and R. Tempone, “Multi-index Monte Carlo: when sparsity meets sampling,” Numer. Math., vol. 132, no. 4, pp. 767–806, 2016

    2016

  13. [21]

    Multilevel and multi-index Monte Carlo methods for the McKean-Vlasov equation,

    A.-L. Haji-Ali and R. Tempone, “Multilevel and multi-index Monte Carlo methods for the McKean-Vlasov equation,” Stat. Comput., vol. 28, no. 4, pp. 923–935, 2018

    2018

  14. [22]

    Sparse grids,

    H.-J. Bungartz and M. Griebel, “Sparse grids,” Acta numerica, vol. 13, no. 1, pp. 147–269, 2004

    2004

  15. [23]

    The parallel finite element system M++ with integrated multilevel preconditioning and multilevel Monte Carlo methods,

    N. Baumgarten and C. Wieners, “The parallel finite element system M++ with integrated multilevel preconditioning and multilevel Monte Carlo methods,” Comput. Math. Appl., vol. 81, pp. 391–406, 2021

    2021

  16. [24]

    A multigrid multilevel Monte Carlo method for transport in the Darcy-Stokes system,

    P. Kumar, P. Luo, F. J. Gaspar, and C. W. Oosterlee, “A multigrid multilevel Monte Carlo method for transport in the Darcy-Stokes system,” J. Comput. Phys., vol. 371, pp. 382–408, 2018

    2018

  17. [25]

    Multilevel Monte Carlo for two phase flow and Buckley-Leverett transport in random heterogeneous porous media,

    F. M¨ uller, P. Jenny, and D. W. Meyer, “Multilevel Monte Carlo for two phase flow and Buckley-Leverett transport in random heterogeneous porous media,” J. Comput. Phys., vol. 250, pp. 685–702, 2013

    2013

  18. [26]

    An explicit link between gaussian fields and gaussian markov random fields: the stochastic partial differential equation approach,

    F. Lindgren, H. Rue, and J. Lindstr¨ om, “An explicit link between gaussian fields and gaussian markov random fields: the stochastic partial differential equation approach,”Journal of the Royal Statistical Society Series B: Statistical Methodology, vol. 73, no. 4, pp. 423–498, 2011

    2011

  19. [27]

    Dirichlet-neumann averaging: The dna of efficient gaussian process simulation,

    R. Kutri and R. Scheichl, “Dirichlet-neumann averaging: The dna of efficient gaussian process simulation,” arXiv preprint arXiv:2412.07929, 2024

    2024 arXiv

  20. [28]

    hp-multilevel monte carlo methods for uncertainty quantification of compressible navier–stokes equations,

    A. Beck, J. D¨ urrw¨ achter, T. Kuhn, F. Meyer, C.-D. Munz, and C. Rohde, “hp-multilevel monte carlo methods for uncertainty quantification of compressible navier–stokes equations,”SIAM Journal on Scientific Computing, vol. 42, no. 4, pp. B1067–B1091, 2020

    2020

  21. [29]

    Flexi: A high order discontinuous galerkin framework for hyperbolic–parabolic con- servation laws,

    N. Krais, A. Beck, T. Bolemann, H. Frank, D. Flad, G. Gassner, F. Hindenlang, M. Hoffmann, T. Kuhn, M. Sonntag, et al., “Flexi: A high order discontinuous galerkin framework for hyperbolic–parabolic con- servation laws,” Computers & Mathematics with Applications, vol. 81, pp. ...

    2021

  22. [30]

    A statistical framework for generating microstructures of two-phase random materials: application to fatigue analysis,

    U. Khristenko, A. Constantinescu, P. L. Tallec, J. T. Oden, and B. Wohlmuth, “A statistical framework for generating microstructures of two-phase random materials: application to fatigue analysis,” Multiscale Modeling & Simulation, vol. 18, no. 1, pp. 21–43, 2020

    2020

  23. [31]

    Finite elements for mat´ ern-type random fields: Uncertainty in computational mechanics and design optimization,

    T. Duswald, B. Keith, B. Lazarov, S. Petrides, and B. Wohlmuth, “Finite elements for mat´ ern-type random fields: Uncertainty in computational mechanics and design optimization,” Computer Methods in Applied Mechanics and Engineering, vol. 429, p. 117146, 2024

    2024

  24. [32]

    A stochastic gradient method with mesh refinement for PDE-constrained optimization under uncertainty,

    C. Geiersbach and W. Wollner, “A stochastic gradient method with mesh refinement for PDE-constrained optimization under uncertainty,” SIAM J. Sci. Comput., vol. 42, no. 5, pp. A2750–A2772, 2020

    2020

  25. [33]

    Complexity analysis of stochastic gradient methods for PDE- constrained optimal control problems with uncertain parameters,

    M. Martin, S. Krumscheid, and F. Nobile, “Complexity analysis of stochastic gradient methods for PDE- constrained optimal control problems with uncertain parameters,” ESAIM Math. Model. Numer. Anal., vol. 55, no. 4, pp. 1599–1633, 2021

    2021

  26. [34]

    Mpp 3.4.1,

    C. Wieners, D. Corallo, D. Schneiderhan, L. Stengel, H. D. N. Pham, and N. Baumgarten, “Mpp 3.4.1,” 2024

    2024

  27. [35]

    Baumgarten, A Fully Parallelized and Budgeted Multi-level Monte Carlo Framework for Partial Differential Equations

    N. Baumgarten, A Fully Parallelized and Budgeted Multi-level Monte Carlo Framework for Partial Differential Equations. PhD thesis, Karlsruher Institut f¨ ur Technologie (KIT), 2023

    2023

  28. [36]

    A filtered multilevel monte carlo method for estimating the expectation of discretized random fields,

    J. Briant, P. Mycek, M. Destouches, O. Goux, S. Gratton, S. G¨ urol, E. Simon, and A. T. Weaver, “A filtered multilevel monte carlo method for estimating the expectation of discretized random fields,” arXiv preprint arXiv:2311.06069, 2023. 26 NIKLAS BAUMGARTEN, ROBERT KUTRI, A...

    2023

  29. [37]

    Application of hierarchical matrices for computing the Karhunen–Lo` eve expansion,

    B. N. Khoromskij, A. Litvinenko, and H. G. Matthies, “Application of hierarchical matrices for computing the Karhunen–Lo` eve expansion,”Computing, vol. 84, no. 1, pp. 49–67, 2009

    2009

  30. [38]

    Karhunen–lo` eve approximation of random fields by generalized fast multipole methods,

    C. Schwab and R. A. Todor, “Karhunen–lo` eve approximation of random fields by generalized fast multipole methods,” Journal of Computational Physics, vol. 217, no. 1, pp. 100–122, 2006

    2006

  31. [39]

    On the low-rank approximation by the pivoted cholesky decomposition,

    H. Harbrecht, M. Peters, and R. Schneider, “On the low-rank approximation by the pivoted cholesky decomposition,” Applied numerical mathematics, vol. 62, no. 4, pp. 428–440, 2012

    2012

  32. [40]

    Fast and exact simulation of stationary Gaussian processes through circulant embedding of the covariance matrix,

    C. R. Dietrich and G. N. Newsam, “Fast and exact simulation of stationary Gaussian processes through circulant embedding of the covariance matrix,” SIAM J. Sci. Comput., vol. 18, no. 4, pp. 1088–1107, 1997

    1997

  33. [41]

    Unified analysis of periodization-based sampling methods for mat´ ern covariances,

    M. Bachmayr, I. G. Graham, V. K. Nguyen, and R. Scheichl, “Unified analysis of periodization-based sampling methods for mat´ ern covariances,”SIAM Journal on Numerical Analysis, vol. 58, no. 5, pp. 2953– 2980, 2020

    2020

  34. [42]

    Efficient white noise sampling and coupling for multilevel monte carlo with nonnested meshes,

    M. Croci, M. B. Giles, M. E. Rognes, and P. E. Farrell, “Efficient white noise sampling and coupling for multilevel monte carlo with nonnested meshes,” SIAM/ASA Journal on Uncertainty Quantification, vol. 6, no. 4, pp. 1630–1655, 2018

    2018

  35. [43]

    Analysis of boundary effects on pde-based sampling of whittle–matern random fields,

    U. Khristenko, L. Scarabosio, P. Swierczynski, E. Ullmann, and B. Wohlmuth, “Analysis of boundary effects on pde-based sampling of whittle–matern random fields,” SIAM/ASA Journal on Uncertainty Quantification, vol. 7, no. 3, pp. 948–974, 2019

    2019

  36. [44]

    Numerical solution of fractional elliptic stochastic pdes with spatial white noise,

    D. Bolin, K. Kirchner, and M. Kov´ acs, “Numerical solution of fractional elliptic stochastic pdes with spatial white noise,” IMA Journal of Numerical Analysis, vol. 40, no. 2, pp. 1051–1073, 2020

    2020

  37. [45]

    Boffi, F

    D. Boffi, F. Brezzi, and M. Fortin, Mixed finite element methods and applications, vol. 44 of Springer Series in Computational Mathematics. Springer, Heidelberg, 2013

    2013

  38. [46]

    Hochbruck, T

    M. Hochbruck, T. Paˇ zur, A. Schulz, E. Thawinan, and C. Wieners, “Efficient time integration for discon- tinuous Galerkin approximations of linear wave equations [Plenary lecture presented at the 83rd Annual GAMM Conference, Darmstadt, 26th–30th March, 2012],” ZAMM Z. Angew. ...

    2012

  39. [47]

    D. A. Di Pietro and A. Ern, Mathematical aspects of discontinuous Galerkin methods, vol. 69 of Math´ ematiques& Applications (Berlin) [Mathematics & Applications]. Springer, Heidelberg, 2012

    2012

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.